Movatterモバイル変換


[0]ホーム

URL:


CN119598453A - Power data back door detection method, device, equipment, storage medium and program product - Google Patents

Power data back door detection method, device, equipment, storage medium and program product
Download PDF

Info

Publication number
CN119598453A
CN119598453ACN202411759788.4ACN202411759788ACN119598453ACN 119598453 ACN119598453 ACN 119598453ACN 202411759788 ACN202411759788 ACN 202411759788ACN 119598453 ACN119598453 ACN 119598453A
Authority
CN
China
Prior art keywords
power data
data
category
cluster
feature
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202411759788.4A
Other languages
Chinese (zh)
Other versions
CN119598453B (en
Inventor
徐文倩
张喜铭
张文瀚
余芸
李鹏
蒋屹新
梁志宏
杨祎巍
毕乐宇
杨秋勇
徐欢
洪超
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China South Power Grid International Co ltd
China Southern Power Grid Co Ltd
Original Assignee
China South Power Grid International Co ltd
China Southern Power Grid Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by China South Power Grid International Co ltd, China Southern Power Grid Co LtdfiledCriticalChina South Power Grid International Co ltd
Priority to CN202411759788.4ApriorityCriticalpatent/CN119598453B/en
Publication of CN119598453ApublicationCriticalpatent/CN119598453A/en
Application grantedgrantedCritical
Publication of CN119598453BpublicationCriticalpatent/CN119598453B/en
Activelegal-statusCriticalCurrent
Anticipated expirationlegal-statusCritical

Links

Classifications

Landscapes

Abstract

Translated fromChinese

一种电力数据后门检测方法、装置、设备、存储介质及程序产品,其方法包括:将待检测电力数据输入到特征提取网络中,得到待检测电力数据的数据特征;特征提取网络中包括多个隐藏层;对数据特征进行特征降维处理,得到低维度数据特征;确定低维度数据特征和预先确定的干净电力数据集对应的多个类别簇分别的基准聚类中心之间的第一欧式距离;基于多个第一欧式距离和预先确定的每个类别簇的距离约束,确定待检测电力数据的检测结果;检测结果为干净电力数据,或中毒电力数据。采用该方法,可减小检测电力数据的计算开销,从而实现对大规模电力数据的实时检测。

A method, device, equipment, storage medium and program product for detecting backdoors in electric power data, the method comprising: inputting the electric power data to be detected into a feature extraction network to obtain data features of the electric power data to be detected; the feature extraction network includes multiple hidden layers; performing feature dimension reduction processing on the data features to obtain low-dimensional data features; determining the first Euclidean distance between the low-dimensional data features and the respective benchmark clustering centers of multiple category clusters corresponding to a predetermined clean electric power data set; determining the detection result of the electric power data to be detected based on the multiple first Euclidean distances and the predetermined distance constraints of each category cluster; the detection result is clean electric power data or poisoned electric power data. The method can reduce the computational overhead of detecting electric power data, thereby realizing real-time detection of large-scale electric power data.

Description

Power data back door detection method, device, equipment, storage medium and program product
Technical Field
The present application relates to the field of computer technologies, and in particular, to a method, an apparatus, a device, a storage medium, and a program product for detecting a power data back door.
Background
The neural network backdoor attack is a malicious attack means aiming at the deep learning model, and an attacker can poison a training data set by injecting part of malicious samples with specific triggers into the training data, so that the model learns strong association between the triggers and target categories in the training process, and a model backdoor is formed. In the reasoning stage, once a poisoning sample with a trigger is input into the model, a back door of the model is activated, and the back door model can generate an error output result according to the intention of an attacker. The back door attack aiming at the power monitoring system is that an attacker falsifies the power system data, a back door trigger is injected into the power data through the modes of voltage zero setting, fake measured value, data noise injection and the like, a poisoning data set is constructed, the back door is hidden implanted in the training process of the power monitoring system, and the power monitoring system makes an error judgment, so that the operation and the social economy of the power system are seriously influenced.
Currently, backdoor defense aiming at the field of power systems mainly comprises a self-coding pretreatment method, a STRIP detection method, a Fine-Pruning method and the like. However, in the above method, when detecting the power data, it is necessary to superimpose a plurality of different modes on the input data to obtain a set of disturbed data samples, and then input a model to evaluate the output change of the set of data samples to detect the poisoning sample. Because each input data needs to be disturbed for multiple times, when the number of input samples is large, a large number of disturbance samples are generated and input into the model, and huge calculation expenditure is caused in the data processing stage and the model reasoning stage, so that the method is not suitable for detecting large-scale real-time power data.
Therefore, how to reduce the calculation overhead of detecting the power data, so as to realize real-time detection of large-scale power data is a problem to be solved.
Disclosure of Invention
The embodiment of the application provides a method, a device, equipment, a storage medium and a program product for detecting a back door of electric power data, which can reduce the calculation cost for detecting the electric power data, thereby realizing the real-time detection of the large-scale electric power data.
In a first aspect, an embodiment of the present application provides a method for detecting power data, including:
Inputting the power data to be detected into a feature extraction network to obtain the data features of the power data to be detected, wherein the feature extraction network comprises a plurality of hidden layers;
Performing feature dimension reduction processing on the data features to obtain low-dimension data features;
Determining a first Euclidean distance between the low-dimensional data characteristic and a reference cluster center of each of a plurality of category clusters corresponding to a predetermined clean power data set;
And determining a detection result of the power data to be detected based on the plurality of first Euclidean distances and the predetermined distance constraint of each class cluster, wherein the detection result is clean power data or toxic power data.
In one embodiment, the determining of the detection result for the power data to be detected based on the plurality of first euclidean distances and the predetermined distance constraint of each category cluster includes determining that the power data to be detected is clean power data when determining that any one of the first euclidean distances is less than or equal to the distance constraint corresponding to the category cluster corresponding to the first euclidean distance, and determining that the power data to be detected is poisoning power data when determining that each of the first euclidean distances is greater than the distance constraint corresponding to the category cluster corresponding to the first euclidean distance.
In one embodiment, reference clustering centers of a plurality of class clusters corresponding to a clean power data set are determined by inputting a plurality of power data in the acquired clean power data set into a feature extraction network to obtain a data feature matrix corresponding to the plurality of power data, performing feature dimension reduction processing on the data feature matrix to obtain a low-dimension feature matrix, clustering the low-dimension feature matrix to obtain a plurality of class clusters, determining an initial clustering center of each class cluster based on each low-dimension feature in each class cluster and the number of low-dimension features in the class cluster, and updating the initial clustering center of each class cluster by a minimized objective function to obtain the reference clustering center of each class cluster.
In one embodiment, the distance constraint for each category cluster is determined by determining a second Euclidean distance between each low dimensional feature in each category cluster and a reference cluster center for the category cluster and determining the distance constraint for each category cluster based on a plurality of second Euclidean distances corresponding to each category cluster.
In one embodiment, determining the distance constraint of each category cluster based on the plurality of second Euclidean distances corresponding to each category cluster comprises determining an average value of the plurality of second Euclidean distances corresponding to the category cluster for each category cluster, and taking the product of the average value and a preset value as the distance constraint of the category cluster.
In one embodiment, performing feature dimension reduction processing on the data features to obtain low-dimensional data features comprises performing feature dimension reduction processing on the data features by using a principal component analysis method to obtain the low-dimensional data features.
In a second aspect, the present application provides a power data back door detection apparatus, the apparatus comprising:
the device comprises a feature extraction module, a feature extraction module and a detection module, wherein the feature extraction module is used for inputting power data to be detected into a feature extraction network to obtain the data features of the power data to be detected;
the feature dimension reduction module is used for carrying out feature dimension reduction processing on the data features to obtain low-dimension data features;
The determining module is used for determining a first Euclidean distance between the low-dimensional data characteristic and a reference cluster center of each of a plurality of category clusters corresponding to a predetermined clean power data set;
The determining module is further used for determining a detection result of the power data to be detected based on the plurality of first Euclidean distances and the predetermined distance constraint of each category cluster, wherein the detection result is clean power data or poisoning power data.
In a third aspect, the present application provides a computer device comprising a memory and a processor, the memory storing a computer program, the processor implementing the following steps when executing the computer program:
Inputting the power data to be detected into a feature extraction network to obtain the data features of the power data to be detected, wherein the feature extraction network comprises a plurality of hidden layers;
Performing feature dimension reduction processing on the data features to obtain low-dimension data features;
Determining a first Euclidean distance between the low-dimensional data characteristic and a reference cluster center of each of a plurality of category clusters corresponding to a predetermined clean power data set;
And determining a detection result of the power data to be detected based on the plurality of first Euclidean distances and the predetermined distance constraint of each class cluster, wherein the detection result is clean power data or toxic power data.
In a fourth aspect, the present application also provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:
Inputting the power data to be detected into a feature extraction network to obtain the data features of the power data to be detected, wherein the feature extraction network comprises a plurality of hidden layers;
Performing feature dimension reduction processing on the data features to obtain low-dimension data features;
Determining a first Euclidean distance between the low-dimensional data characteristic and a reference cluster center of each of a plurality of category clusters corresponding to a predetermined clean power data set;
And determining a detection result of the power data to be detected based on the plurality of first Euclidean distances and the predetermined distance constraint of each class cluster, wherein the detection result is clean power data or toxic power data.
In a fifth aspect, the application also provides a computer program product comprising a computer program which, when executed by a processor, performs the steps of:
Inputting the power data to be detected into a feature extraction network to obtain the data features of the power data to be detected, wherein the feature extraction network comprises a plurality of hidden layers;
Performing feature dimension reduction processing on the data features to obtain low-dimension data features;
Determining a first Euclidean distance between the low-dimensional data characteristic and a reference cluster center of each of a plurality of category clusters corresponding to a predetermined clean power data set;
And determining a detection result of the power data to be detected based on the plurality of first Euclidean distances and the predetermined distance constraint of each class cluster, wherein the detection result is clean power data or toxic power data.
The power data back door detection method, the device, the equipment, the storage medium and the program product can enable computer equipment to input power data to be detected into a feature extraction network to obtain data features of the power data to be detected, wherein the feature extraction network comprises a plurality of hidden layers, feature dimension reduction processing is conducted on the data features to obtain low-dimension data features, first Euclidean distances between the low-dimension data features and reference cluster centers of a plurality of category clusters corresponding to a predetermined clean power data set are determined, detection results of the power data to be detected are determined based on the first Euclidean distances and distance constraint of each category cluster, and the detection results are clean power data or toxic power data. By adopting the method, the computer equipment can obtain the low-dimensional deep data characteristic by extracting the deep data characteristic of the electric power data to be detected and performing dimension reduction processing on the deep data characteristic, then determine the Euclidean distance between the low-dimensional deep data characteristic and the reference cluster center of each class cluster corresponding to the predetermined clean electric power data set, judge whether the distance constraint is met, and further determine the detection result aiming at the electric power data to be detected. Compared with the existing power data detection method, the method has the advantages that complicated data processing is not needed for input data, so that the calculation cost for detecting the power data can be reduced, and further real-time detection of large-scale power data is realized.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the related art, the drawings that are needed in the description of the embodiments of the present application or the related technologies will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and other related drawings may be obtained according to these drawings without inventive effort to those of ordinary skill in the art.
Fig. 1 is a schematic diagram of an application scenario of a power data back door detection method according to an embodiment of the present application;
fig. 2 is a schematic flow chart of a method for detecting a back door of power data according to an embodiment of the present application;
FIG. 3 is a flowchart of another power data back door detection method according to an embodiment of the present application;
Fig. 4 is a schematic structural diagram of a power data back door detection device according to an embodiment of the present application;
Fig. 5 is a schematic structural diagram of a computer device according to an embodiment of the present application.
Detailed Description
The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.
The application scenario of the power data detection method provided by the embodiment of the application is described below.
Referring to fig. 1, fig. 1 is a schematic diagram of an application scenario of a power data back door detection method according to an embodiment of the present application. As shown in fig. 1, the system includes a computer device 101 (the computer device 101 is shown as a terminal device in fig. 1 as an example) and a database server 102, wherein data transmission between the computer device 101 and the database server 102 is performed via a network.
The database server 102 may be configured to store the power data to be detected, a reference cluster center corresponding to a predetermined clean power data set and a distance constraint of each class cluster.
The computer device 101 may obtain the power data to be detected from the database server 102, input the power data to be detected into a feature extraction network including a plurality of hidden layers to obtain deep data features of the power data to be detected, perform feature dimension reduction on the deep data features to obtain low-dimensional deep data features of the power data to be detected, obtain reference cluster centers of a plurality of category clusters corresponding to a predetermined clean power data set from the database server 102, determine euclidean distances between the low-dimensional deep data features and the reference cluster centers of each category cluster, obtain distance constraints of each category cluster from the database server 102, determine a detection result of the power data to be detected based on the plurality of euclidean distances and the distance constraints of each category cluster, and obtain the detection result as the clean power data or the toxic power data. By adopting the method, the computer equipment can obtain the low-dimensional deep data characteristic by extracting the deep data characteristic of the electric power data to be detected and performing dimension reduction processing on the deep data characteristic, then determine the Euclidean distance between the low-dimensional deep data characteristic and the reference cluster center of each class cluster corresponding to the predetermined clean electric power data set, judge whether the distance constraint is met, and further determine the detection result aiming at the electric power data to be detected. Compared with the existing power data detection method, the method has the advantages that complicated data processing is not needed for input data, so that the calculation cost for detecting the power data can be reduced, and further real-time detection of large-scale power data is realized.
Alternatively, the computer device 101 may be a terminal device or a server. Among them, the terminal devices mentioned herein may include, but are not limited to, smart phones, tablet computers, notebook computers, desktop computers, smart watches, smart televisions, smart car terminals, etc. The server mentioned here may be a separate physical server, or may be a server cluster or a distributed system formed by a plurality of physical servers.
Referring to fig. 2, fig. 2 is a flow chart of a method for detecting a back door of power data according to an embodiment of the application. The method may be performed by a computer device, such as the computer device 101 described above. As shown in fig. 2, the power data back door detection method may include, but is not limited to, the following steps:
S201, inputting the power data to be detected into a feature extraction network to obtain the data features of the power data to be detected, wherein the feature extraction network comprises a plurality of hidden layers.
Because the feature extraction network comprises a plurality of hidden layers, the computer equipment inputs the power data to be detected into the feature extraction network, and the obtained data features of the power data to be detected are deep data features. The deep data features contain richer semantic information, which is more beneficial to distinguishing clean power data from poisoning power data. Therefore, the accuracy of the detection result can be improved when the electric power data to be detected is detected based on the deep data characteristics.
And S202, performing feature dimension reduction processing on the data features to obtain low-dimension data features.
In this way, the computer device can obtain the low-dimensional data feature by reducing the dimension of the data feature, thereby being beneficial to reducing the calculation cost of detecting the power data.
S203, determining a first Euclidean distance between the low-dimensional data characteristic and a reference cluster center of each of a plurality of category clusters corresponding to a predetermined clean power data set.
In an alternative embodiment, before step S203, the computer device may further determine a reference cluster center of each of the plurality of category clusters corresponding to the clean power data set.
S204, determining a detection result of the power data to be detected based on a plurality of first Euclidean distances and a predetermined distance constraint of each category cluster, wherein the detection result is clean power data or poisoning power data.
In an alternative embodiment, prior to step S204, the computer device may further determine respective distance constraints for a plurality of category clusters corresponding to the clean power dataset.
In the embodiment of the application, the computer equipment can input the electric power data to be detected into the characteristic extraction network to obtain the data characteristic of the electric power data to be detected, the characteristic extraction network comprises a plurality of hidden layers, the characteristic dimension reduction processing is carried out on the data characteristic to obtain the low-dimension data characteristic, the first Euclidean distance between the low-dimension data characteristic and the reference clustering center of each of a plurality of category clusters corresponding to a predetermined clean electric power data set is determined, the detection result of the electric power data to be detected is determined based on the first Euclidean distances and the distance constraint of each category cluster, and the detection result is clean electric power data or toxic electric power data. By adopting the method, the computer equipment can obtain the low-dimensional deep data characteristic by extracting the deep data characteristic of the electric power data to be detected and performing dimension reduction processing on the deep data characteristic, then determine the Euclidean distance between the low-dimensional deep data characteristic and the reference cluster center of each class cluster corresponding to the predetermined clean electric power data set, judge whether the distance constraint is met, and further determine the detection result aiming at the electric power data to be detected. Compared with the existing power data detection method, the method has the advantages that complicated data processing is not needed for input data, so that the calculation cost for detecting the power data can be reduced, and further real-time detection of large-scale power data is realized.
Referring to fig. 3, fig. 3 is a flowchart illustrating another power data back door detection method according to an embodiment of the application. In comparison with the power data back door detection method shown in fig. 2, the method shown in fig. 3 further illustrates how the computer device determines the reference cluster center of each of the plurality of class clusters corresponding to the clean power data set, and how to determine the distance constraint of each class cluster. As shown in fig. 3, the power data back door detection method may include, but is not limited to, the following steps:
s301, acquiring a clean power data set.
In defending against a back door attack of a power system, a computer device may acquire a clean power data set and verify the basic performance of the power system based on the clean power data set, assuming that an defender cannot access the original training power data set, but can access a set of clean verification power data sets (or become the clean power data set).
S302, inputting a plurality of pieces of power data in the obtained clean power data set into a feature extraction network to obtain a data feature matrix corresponding to the plurality of pieces of power data, wherein the feature extraction network comprises a plurality of hidden layers.
Because the feature extraction network comprises a plurality of hidden layers, the computer equipment inputs a plurality of pieces of power data in the clean power data set into the feature extraction network, and the obtained data feature matrix of the plurality of pieces of power data comprises deep data features of each piece of power data in the plurality of pieces of power data.
S303, performing feature dimension reduction processing on the data feature matrix to obtain a low-dimension feature matrix.
In an alternative embodiment, the computer device performs feature dimension reduction processing on the data feature matrix to obtain a low-dimensional feature matrix, which may include performing feature dimension reduction processing on the data feature matrix by using a principal component analysis method to obtain the low-dimensional feature matrix.
The principal component analysis method (PRINCIPAL COMPONENT ANALYSIS, PCA) is also called principal component analysis, and aims to convert multiple indexes into a few comprehensive indexes (i.e. principal components) by using the idea of dimension reduction, wherein each principal component can reflect most of information of an original variable, and the contained information is not repeated. The method introduces various variables and simultaneously attributes complex factors into a plurality of main components, so that the problems are simplified, and the obtained result is more scientific and effective data information.
The process of obtaining a low-dimensional feature matrix by performing feature dimension reduction processing on a data feature matrix (denoted as fl) by using a principal component analysis method by using computer equipment is described below.
Step one, determining a covariance matrix of the data feature matrix fl.
And secondly, decomposing the eigenvalue of the data eigenvalue matrix fl through a covariance matrix, and determining the main component in fl.
And thirdly, selecting m principal components with the largest reservation, and projecting the original data characteristic fl onto the m principal components so as to form a matrix P= (P1,P2,...,Pm).
And fourthly, processing the matrix P to obtain a plurality of low-dimensional feature matrices Zpca corresponding to the electric power data.
Optionally, when the computer device processes the matrix P to obtain the low-dimensional feature matrix Zpca corresponding to the plurality of electric power data, the following formula (1) may be adopted.
(1)
In the formula (1), Z represents a data normalization result of the data feature matrix fl.
In the application, the computer equipment performs dimension reduction processing on the deep data feature matrix of the electric power data, maps the high-dimensional features into the low-dimensional space, can enable the k-means cluster to more effectively distinguish different types of data, and reduces noise existing in the high-dimensional data features.
S304, clustering the low-dimensional feature matrix to obtain a plurality of category clusters.
In an alternative embodiment, the computer device may use a K-means clustering method to cluster multiple low-dimensional features to obtain multiple class clusters.
Each class cluster contains one or more low-dimensional characteristics of clean power data, and the clean power data corresponding to the low-dimensional characteristics in each class cluster has the same or similar properties. In addition, each class cluster has a cluster center for representing the average point of the class cluster.
S305, determining an initial clustering center of each category cluster based on each low-dimensional feature in each category cluster and the number of low-dimensional features in the category cluster.
In an alternative embodiment, the computer device may employ the following equation (2) when determining the initial cluster center for each category cluster based on each low-dimensional feature in each category cluster and the number of low-dimensional features in that category cluster.
(2)
In formula (2), Cj represents a set of jth category clusters; The number of data points (low-dimensional features) belonging to the j-th class cluster is represented. Zi represents the data points (low-dimensional features) assigned to the jth class cluster (i.e., Cj), and Zi represents the initial cluster center of the jth class cluster from the previously determined low-dimensional feature matrix Zpca;uj.
S306, updating the initial cluster center of each category cluster through the minimized objective function to obtain the reference cluster center of each category cluster.
In an alternative embodiment, the objective function may be as shown in equation (3) below.
(3)
In equation (3), C represents the total number of category clusters, Cj represents the set of j-th category clusters, zi represents the data points (low-dimensional features) assigned to the j-th category cluster (i.e., Cj), uj represents the initial cluster center of the j-th category cluster, and E represents the objective function.
Optionally, the computer device, by minimizing equation (3) above, determining a reference cluster center (denoted as) Then, the reference cluster centers corresponding to all the category clusters can be determinedWherein C represents the total number of category clusters.
S307, determining a second Euclidean distance between each low dimensional feature in each category cluster and the reference cluster center of the category cluster.
In an alternative embodiment, the computer device may employ the following equation (4) in determining the second Euclidean distance between each low dimensional feature in each category cluster and the reference cluster center of that category cluster.
(4)
In equation (4), zi represents the i-th low-dimensional feature assigned to the j-th class cluster; The reference cluster center of the jth category cluster is represented, and di represents the Euclidean distance between the ith low-dimensional feature in the jth category cluster and the reference cluster center of the jth category cluster.
S308, determining the distance constraint of each category cluster based on a plurality of second Euclidean distances corresponding to each category cluster.
In an alternative embodiment, the computer device determines a distance constraint of each category cluster based on the plurality of second Euclidean distances corresponding to each category cluster, and the method can include determining, for each category cluster, an average value of the plurality of second Euclidean distances corresponding to the category cluster, and taking the product of the average value and a preset value as the distance constraint of the category cluster.
In this embodiment, when the computer device determines, for each category cluster, an average value of the plurality of second euclidean distances corresponding to the category cluster, the following formula (5) may be adopted.
(5)
In the formula (5) of the present invention,The average value of the Euclidean distances corresponding to the jth category cluster is represented, di is represented by the Euclidean distance between the ith low dimensional feature in the jth category cluster and the reference cluster center of the jth category cluster, and the physical meaning of other letters can refer to the explanation of the physical meaning of the letters in the formula (4) above, and the description is omitted here.
In this embodiment, the preset value is, for example, 3. Wherein 3 is an empirical value based on multiple experiments. In the case of a preset value of 3, the computer device may determine that the distance constraint of the jth category cluster is
S309, inputting the power data to be detected into a feature extraction network to obtain the data features of the power data to be detected.
In an alternative embodiment, the description of step S309 may be referred to in the foregoing description of step S301, and will not be repeated here.
And S310, performing feature dimension reduction processing on the data features by using a principal component analysis method to obtain low-dimensional data features.
S311, determining a first Euclidean distance between the low-dimensional data characteristic and a reference cluster center of each of a plurality of category clusters corresponding to a predetermined clean power data set.
In an alternative embodiment, the computer device may calculate reference cluster centers for all class clusters corresponding to the low-dimensional data features and the clean power data setA first euclidean distance between each of the reference cluster centers.
S312, determining a detection result of the power data to be detected based on the plurality of first Euclidean distances and the predetermined distance constraint of each class cluster, wherein the detection result is clean power data or poisoning power data.
In an alternative embodiment, the computer device determines the detection result of the power data to be detected based on the plurality of first euclidean distances and the predetermined distance constraint of each category cluster, and may include determining that the power data to be detected is clean power data if any one of the first euclidean distances is determined to be less than or equal to the distance constraint corresponding to the category cluster corresponding to the first euclidean distance, and determining that the power data to be detected is toxic power data if each of the first euclidean distances is determined to be greater than the distance constraint corresponding to the category cluster corresponding to the first euclidean distance.
For example, assume that among a plurality of first euclidean distances, a first euclidean distance 1 is a euclidean distance between a low dimensional data feature of power data to be detected and a reference cluster center of a1 st category cluster in a predetermined clean power data set, the first euclidean distance 1 is 0.18, and a distance constraint of the 1 st category cluster is=0.21. In this case, the computer device may determine that the first euclidean distance 1 (0.18) is less than the distance constraint (0.21) of the 1 st category cluster, at which point the computer device may determine that the power data to be detected is clean power data.
For another example, assume that the plurality of category clusters includes a category cluster 1, a category cluster 2, and a category cluster 3, wherein a distance constraint of the category cluster 1 is 0.21, a distance constraint of the category cluster 2 is 0.24, a distance constraint of the category cluster 3 is 0.18, and a plurality of first euclidean distances include a first euclidean distance 1, a first euclidean distance 2, and a first euclidean distance 3, wherein the first euclidean distance 1 is a euclidean distance between a low dimensional data feature of the power data to be detected and a reference cluster center of the category cluster 1, the first euclidean distance 1 is 0.22, the first euclidean distance 2 is a euclidean distance between a low dimensional data feature of the power data to be detected and a reference cluster center of the category cluster 2, the first euclidean distance 2 is 0.26, and the first euclidean distance 3 is a euclidean distance between a low dimensional data feature of the power data to be detected and a reference cluster center of the category cluster 2. In this case, the computer device may determine that the first euclidean distance 1 (0.22) is greater than the distance constraint (0.21) of the category cluster 1, the first euclidean distance 2 (0.26) is greater than the distance constraint (0.24) of the category cluster 2, and the first euclidean distance 3 (0.20) is greater than the distance constraint (0.18) of the category cluster 3, at which point the computer device may determine that the power data to be detected is poisoning power data.
In the embodiment of the application, the computer equipment can acquire a clean power data set, determine the reference cluster centers of a plurality of class clusters corresponding to the clean power data set, then determine the Euclidean distance between all samples in each class cluster and the reference cluster center of the class cluster, determine the distance constraint of each class cluster based on the Euclidean distances, then determine the low-dimensional data characteristic of the power data to be detected in the process of detecting the power data to be detected, determine the Euclidean distance between the low-dimensional data characteristic and the reference cluster center of each class cluster, and finally determine the power data to be detected as clean power data or as poisoning power data based on the Euclidean distances and the distance constraint of each class cluster. Compared with the existing power data detection method, the method has the advantages that complicated data processing is not needed for input data, so that the calculation cost for detecting the power data can be reduced, and further real-time detection of large-scale power data is realized.
In addition, the application utilizes the feature extraction network comprising a plurality of hidden layers to extract the features of the electric data, and can extract the deep features of the electric data (or utilize the deep features of the electric data output by the last hidden layer in the feature extraction network), then, the deep features are used as the basis of back door analysis, and as the deep features are used as the highest-order features of the input electric data, the decision process of the model is determined, and meanwhile, the decision process of how to correctly identify clean samples by the back door model is implied, and the reasons for incorrectly predicting the back door samples as target types are implied, therefore, the deep features are analyzed, the accuracy of detecting the electric data to be detected can be improved, or the poisoning electric data and the clean electric data can be distinguished more accurately.
It should be understood that, although the steps in the flowcharts related to the embodiments described above are sequentially shown as indicated by arrows, these steps are not necessarily sequentially performed in the order indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in the flowcharts described in the above embodiments may include a plurality of steps or a plurality of stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily performed sequentially, but may be performed alternately or alternately with at least some of the other steps or stages.
Based on the same inventive concept, the embodiment of the application also provides a power data back door detection device for realizing the power data back door detection method. The implementation of the solution provided by the device is similar to the implementation described in the above method, so the specific limitation in the embodiments of the device for detecting a back door of electric power data provided below may refer to the limitation of the method for detecting a back door of electric power data hereinabove, and will not be repeated here.
Referring to fig. 4, fig. 4 is a schematic structural diagram of a power data back door detection device according to an embodiment of the application. As shown in fig. 4, the power data detection apparatus may include, but is not limited to:
the feature extraction module 401 is configured to input power data to be detected into a feature extraction network to obtain a data feature of the power data to be detected;
the feature dimension reduction module 402 is configured to perform feature dimension reduction processing on the data feature to obtain a low-dimension data feature;
A determining module 403, configured to determine a first euclidean distance between the low dimensional data feature and a reference cluster center of each of a plurality of category clusters corresponding to a predetermined clean power data set;
The determining module 403 is further configured to determine a detection result of the power data to be detected based on the plurality of first euclidean distances and a predetermined distance constraint of each category cluster, where the detection result is clean power data or toxic power data.
In one embodiment, the determining module 403 is configured to determine, when determining a detection result for the power data to be detected based on the plurality of first euclidean distances and the predetermined distance constraint of each class cluster, specifically configured to determine that the power data to be detected is clean power data if any one of the first euclidean distances is less than or equal to the distance constraint corresponding to the class cluster corresponding to the first euclidean distance, and determine that the power data to be detected is toxic power data if each of the first euclidean distances is greater than the distance constraint corresponding to the class cluster corresponding to the first euclidean distance.
In one embodiment, the feature extraction module 401 is further configured to input a plurality of pieces of power data in the obtained clean power data set into the feature extraction network to obtain a data feature matrix corresponding to the plurality of pieces of power data, the feature dimension reduction module 402 is further configured to perform feature dimension reduction processing on the data feature matrix to obtain a low-dimensional feature matrix, cluster the low-dimensional feature matrix to obtain a plurality of class clusters, the determination module 403 is further configured to determine an initial cluster center of each class cluster based on each low-dimensional feature in each class cluster and the number of low-dimensional features in the class cluster, and update the initial cluster center of each class cluster by minimizing an objective function to obtain a reference cluster center of each class cluster.
In one embodiment, the determining module 403 is further configured to determine a second euclidean distance between each low dimensional feature in each category cluster and the reference cluster center of the category cluster, and determine a distance constraint for each category cluster based on the plurality of second euclidean distances corresponding to each category cluster.
In one embodiment, the determining module 403 is configured to determine, for each category cluster, an average value of the plurality of second euclidean distances corresponding to the category cluster, and taking a product of the average value and a preset value as the distance constraint of the category cluster when determining the distance constraint of each category cluster based on the plurality of second euclidean distances corresponding to the category cluster.
In one embodiment, the feature dimension reduction module 402 is configured to perform feature dimension reduction processing on the data feature to obtain a low-dimensional data feature, and is specifically configured to perform feature dimension reduction processing on the data feature by using a principal component analysis method to obtain the low-dimensional data feature.
The above-mentioned various modules in the power data back door detection apparatus may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or independent of a processor in the terminal device, or may be stored in software in a memory in the terminal device, so that the processor may call and execute operations corresponding to the above modules.
In an exemplary embodiment, an embodiment of the present application provides a computer device, which may be a terminal, and an internal structure diagram thereof may be as shown in fig. 5. The computer device includes a processor, a memory, an input/output interface, a communication interface, a display unit, and an input means. The processor, the memory and the input/output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input/output interface. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input/output interface of the computer device is used to exchange information between the processor and the external device. The Communication interface of the computer device is used for conducting wired or wireless Communication with an external terminal, and the wireless Communication can be realized through WIFI, a mobile cellular network, near field Communication (NEAR FIELD Communication) or other technologies. The computer program when executed by a processor implements a power data back door detection method.
It will be appreciated by those skilled in the art that the structure shown in FIG. 5 is merely a block diagram of some of the structures associated with the present inventive arrangements and is not limiting of the computer device to which the present inventive arrangements may be applied, and that a particular computer device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.
In one exemplary embodiment, the application provides a computer device comprising a memory and a processor, the memory having stored therein a computer program which when executed by the processor performs the steps of:
Inputting the power data to be detected into a feature extraction network to obtain the data features of the power data to be detected, wherein the feature extraction network comprises a plurality of hidden layers;
Performing feature dimension reduction processing on the data features to obtain low-dimension data features;
Determining a first Euclidean distance between the low-dimensional data characteristic and a reference cluster center of each of a plurality of category clusters corresponding to a predetermined clean power data set;
And determining a detection result of the power data to be detected based on the plurality of first Euclidean distances and the predetermined distance constraint of each class cluster, wherein the detection result is clean power data or toxic power data.
In one embodiment, when the processor executes the computer program to determine a detection result for the power data to be detected based on a plurality of first euclidean distances and a predetermined distance constraint of each category cluster, the method specifically includes determining that the power data to be detected is clean power data if any one of the first euclidean distances is determined to be less than or equal to a distance constraint corresponding to the category cluster corresponding to the first euclidean distance, and determining that the power data to be detected is toxic power data if each of the first euclidean distances is determined to be greater than the distance constraint corresponding to the category cluster corresponding to the first euclidean distance.
In one embodiment, the processor executes the computer program to further perform the steps of inputting a plurality of pieces of power data in the obtained clean power data set into the feature extraction network to obtain a data feature matrix corresponding to the plurality of pieces of power data, performing feature dimension reduction processing on the data feature matrix to obtain a low-dimensional feature matrix, clustering the low-dimensional feature matrix to obtain a plurality of category clusters, determining an initial cluster center of each category cluster based on each low-dimensional feature in each category cluster and the number of the low-dimensional features in the category cluster, and updating the initial cluster center of each category cluster by minimizing an objective function to obtain a reference cluster center of each category cluster.
In one embodiment, the processor executing the computer program further implements the steps of determining a second Euclidean distance between each low dimensional feature in each category cluster and the reference cluster center of the category cluster, and determining a distance constraint for each category cluster based on a plurality of second Euclidean distances corresponding to each category cluster.
In one embodiment, when the processor executes the computer program to determine the distance constraint of each category cluster based on the plurality of second Euclidean distances corresponding to each category cluster, the method specifically comprises the steps of determining, for each category cluster, an average value of the plurality of second Euclidean distances corresponding to the category cluster, and taking the product of the average value and a preset value as the distance constraint of the category cluster.
In one embodiment, when the processor executes the computer program to realize the feature dimension reduction processing on the data features to obtain the low-dimension data features, the processor specifically realizes the steps of utilizing a principal component analysis method to perform the feature dimension reduction processing on the data features to obtain the low-dimension data features.
In one exemplary embodiment, the application provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:
Inputting the power data to be detected into a feature extraction network to obtain the data features of the power data to be detected, wherein the feature extraction network comprises a plurality of hidden layers;
Performing feature dimension reduction processing on the data features to obtain low-dimension data features;
Determining a first Euclidean distance between the low-dimensional data characteristic and a reference cluster center of each of a plurality of category clusters corresponding to a predetermined clean power data set;
And determining a detection result of the power data to be detected based on the plurality of first Euclidean distances and the predetermined distance constraint of each class cluster, wherein the detection result is clean power data or toxic power data.
In one embodiment, when the computer program is executed by the processor to determine a detection result for the power data to be detected based on the plurality of first euclidean distances and the predetermined distance constraint of each category cluster, the method specifically includes determining that the power data to be detected is clean power data if any one of the first euclidean distances is determined to be less than or equal to the distance constraint corresponding to the category cluster corresponding to the first euclidean distance, and determining that the power data to be detected is toxic power data if each of the first euclidean distances is determined to be greater than the distance constraint corresponding to the category cluster corresponding to the first euclidean distance.
In one embodiment, the computer program when executed by the processor further comprises the steps of inputting a plurality of pieces of power data in the obtained clean power data set into the feature extraction network to obtain data feature matrixes corresponding to the plurality of pieces of power data, performing feature dimension reduction processing on the data feature matrixes to obtain low-dimensional feature matrixes, clustering the low-dimensional feature matrixes to obtain a plurality of category clusters, determining an initial cluster center of each category cluster based on each low-dimensional feature in each category cluster and the number of the low-dimensional features in the category cluster, and updating the initial cluster center of each category cluster by a minimized objective function to obtain a reference cluster center of each category cluster.
In one embodiment, the computer program when executed by the processor further performs the steps of determining a second Euclidean distance between each low dimensional feature in each category cluster and a reference cluster center of the category cluster, and determining a distance constraint for each category cluster based on a plurality of second Euclidean distances corresponding to each category cluster.
In one embodiment, when the computer program is executed by the processor to determine the distance constraint of each category cluster based on the plurality of second euclidean distances corresponding to each category cluster, the method specifically includes determining, for each category cluster, an average value of the plurality of second euclidean distances corresponding to the category cluster, and taking a product of the average value and a preset value as the distance constraint of the category cluster.
In one embodiment, when the computer program is executed by the processor to perform feature dimension reduction processing on the data features to obtain low-dimensional data features, the method specifically includes the steps of performing feature dimension reduction processing on the data features by using a principal component analysis method to obtain the low-dimensional data features.
In one exemplary embodiment, the application provides a computer program product comprising a computer program which, when executed by a processor, performs the steps of:
Inputting the power data to be detected into a feature extraction network to obtain the data features of the power data to be detected, wherein the feature extraction network comprises a plurality of hidden layers;
Performing feature dimension reduction processing on the data features to obtain low-dimension data features;
Determining a first Euclidean distance between the low-dimensional data characteristic and a reference cluster center of each of a plurality of category clusters corresponding to a predetermined clean power data set;
And determining a detection result of the power data to be detected based on the plurality of first Euclidean distances and the predetermined distance constraint of each class cluster, wherein the detection result is clean power data or toxic power data.
In one embodiment, when the computer program is executed by the processor to determine a detection result for the power data to be detected based on the plurality of first euclidean distances and the predetermined distance constraint of each category cluster, the method specifically includes determining that the power data to be detected is clean power data if any one of the first euclidean distances is determined to be less than or equal to the distance constraint corresponding to the category cluster corresponding to the first euclidean distance, and determining that the power data to be detected is toxic power data if each of the first euclidean distances is determined to be greater than the distance constraint corresponding to the category cluster corresponding to the first euclidean distance.
In one embodiment, the computer program when executed by the processor further comprises the steps of inputting a plurality of pieces of power data in the obtained clean power data set into the feature extraction network to obtain data feature matrixes corresponding to the plurality of pieces of power data, performing feature dimension reduction processing on the data feature matrixes to obtain low-dimensional feature matrixes, clustering the low-dimensional feature matrixes to obtain a plurality of category clusters, determining an initial cluster center of each category cluster based on each low-dimensional feature in each category cluster and the number of the low-dimensional features in the category cluster, and updating the initial cluster center of each category cluster by a minimized objective function to obtain a reference cluster center of each category cluster.
In one embodiment, the computer program when executed by the processor further performs the steps of determining a second Euclidean distance between each low dimensional feature in each category cluster and a reference cluster center of the category cluster, and determining a distance constraint for each category cluster based on a plurality of second Euclidean distances corresponding to each category cluster.
In one embodiment, when the computer program is executed by the processor to determine the distance constraint of each category cluster based on the plurality of second euclidean distances corresponding to each category cluster, the method specifically includes determining, for each category cluster, an average value of the plurality of second euclidean distances corresponding to the category cluster, and taking a product of the average value and a preset value as the distance constraint of the category cluster.
In one embodiment, when the computer program is executed by the processor to perform feature dimension reduction processing on the data features to obtain low-dimensional data features, the method specifically includes the steps of performing feature dimension reduction processing on the data features by using a principal component analysis method to obtain the low-dimensional data features.
It should be noted that, the data (including but not limited to the power data to be detected, the clean power data set, etc.) related to the present application are all information and data authorized by the user or fully authorized by each party, and the collection, use and processing of the related data are required to meet the related regulations.
Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, database, or other medium used in embodiments provided herein may include at least one of non-volatile memory and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high density embedded nonvolatile Memory, resistive random access Memory (RESISTIVE RANDOM ACCESS MEMORY, reRAM), magneto-resistive Memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric Memory (Ferroelectric Random Access Memory, FRAM), phase change Memory (PHASE CHANGE Memory, PCM), graphene Memory, and the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, and the like. By way of illustration, and not limitation, RAM can be in various forms such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc. The databases referred to in the embodiments provided herein may include at least one of a relational database and a non-relational database. The non-relational database may include, but is not limited to, a blockchain-based distributed database, and the like. The processor referred to in the embodiments provided in the present application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computation, an artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) processor, or the like, but is not limited thereto.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the present application.
The foregoing examples illustrate only a few embodiments of the application and are described in detail herein without thereby limiting the scope of the application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of the application should be assessed as that of the appended claims.

Claims (10)

Translated fromChinese
1.一种电力数据后门检测方法,其特征在于,所述方法包括:1. A method for detecting a backdoor in power data, characterized in that the method comprises:将待检测电力数据输入到特征提取网络中,得到所述待检测电力数据的数据特征;所述特征提取网络中包括多个隐藏层;Inputting the power data to be detected into a feature extraction network to obtain data features of the power data to be detected; the feature extraction network includes multiple hidden layers;对所述数据特征进行特征降维处理,得到低维度数据特征;Performing feature dimensionality reduction processing on the data features to obtain low-dimensional data features;确定所述低维度数据特征和预先确定的干净电力数据集对应的多个类别簇分别的基准聚类中心之间的第一欧式距离;Determine a first Euclidean distance between the low-dimensional data feature and respective benchmark cluster centers of a plurality of category clusters corresponding to a predetermined clean power data set;基于多个所述第一欧式距离和预先确定的每个所述类别簇的距离约束,确定所述待检测电力数据的检测结果;所述检测结果为干净电力数据,或中毒电力数据。Based on the plurality of the first Euclidean distances and the predetermined distance constraint of each of the category clusters, a detection result of the power data to be detected is determined; the detection result is clean power data or poisoned power data.2.根据权利要求1所述的方法,其特征在于,所述基于多个所述第一欧式距离和预先确定的每个所述类别簇的距离约束,确定针对所述待检测电力数据的检测结果,包括:2. The method according to claim 1, characterized in that the step of determining the detection result for the power data to be detected based on the plurality of the first Euclidean distances and the predetermined distance constraint of each of the category clusters comprises:在确定任意一个所述第一欧式距离小于等于该第一欧式距离对应的类别簇所对应的距离约束的情况下,确定所述待检测电力数据为干净电力数据;In the case where it is determined that any one of the first Euclidean distances is less than or equal to the distance constraint corresponding to the category cluster corresponding to the first Euclidean distance, determining that the power data to be detected is clean power data;在确定每个所述第一欧式距离均大于该第一欧式距离对应的类别簇所对应的距离约束的情况下,确定所述待检测电力数据为中毒电力数据。When it is determined that each of the first Euclidean distances is greater than the distance constraint corresponding to the category cluster corresponding to the first Euclidean distance, it is determined that the power data to be detected is poisoned power data.3.根据权利要求1所述的方法,其特征在于,所述干净电力数据集对应的多个类别簇分别的基准聚类中心是通过以下方式确定的:3. The method according to claim 1, characterized in that the benchmark cluster centers of the multiple category clusters corresponding to the clean power data set are determined by:将获取到的干净电力数据集中的多个电力数据输入到特征提取网络中,得到多个所述电力数据对应的数据特征矩阵;Inputting multiple power data in the acquired clean power data set into a feature extraction network to obtain a data feature matrix corresponding to the multiple power data;对所述数据特征矩阵进行特征降维处理,得到低维度特征矩阵;Performing feature dimensionality reduction processing on the data feature matrix to obtain a low-dimensional feature matrix;对所述低维度特征矩阵进行聚类,得到多个类别簇;Clustering the low-dimensional feature matrix to obtain multiple category clusters;基于每个所述类别簇中的每个低维度特征和该类别簇中所述低维度特征的数量,确定每个所述类别簇的初始聚类中心;Determine an initial cluster center of each of the category clusters based on each low-dimensional feature in each of the category clusters and the number of the low-dimensional features in the category cluster;通过最小化目标函数对每个所述类别簇的初始聚类中心进行更新,得到每个所述类别簇的基准聚类中心。The initial cluster center of each of the category clusters is updated by minimizing the objective function to obtain the benchmark cluster center of each of the category clusters.4.根据权利要求3所述的方法,其特征在于,所述每个所述类别簇的距离约束是通过以下方式确定的:4. The method according to claim 3, characterized in that the distance constraint of each of the category clusters is determined by:确定每个所述类别簇中的每个所述低维度特征到该类别簇的基准聚类中心之间的第二欧式距离;Determine a second Euclidean distance between each of the low-dimensional features in each of the category clusters and a reference cluster center of the category cluster;基于每个所述类别簇对应的多个所述第二欧式距离,确定每个所述类别簇的距离约束。Based on the plurality of the second Euclidean distances corresponding to each of the category clusters, a distance constraint of each of the category clusters is determined.5.根据权利要求4所述的方法,其特征在于,所述基于每个所述类别簇对应的多个所述第二欧式距离,确定每个所述类别簇的距离约束,包括:5. The method according to claim 4, characterized in that the determining the distance constraint of each of the category clusters based on the plurality of the second Euclidean distances corresponding to each of the category clusters comprises:针对每个所述类别簇,确定该类别簇对应的多个所述第二欧式距离的平均值;For each of the category clusters, determining an average value of a plurality of the second Euclidean distances corresponding to the category cluster;将所述平均值和预设数值的乘积,作为该类别簇的距离约束。The product of the average value and a preset value is used as the distance constraint of the category cluster.6.根据权利要求1至5任一项所述的方法,其特征在于,所述对所述数据特征进行特征降维处理,得到低维度数据特征,包括:6. The method according to any one of claims 1 to 5, characterized in that the step of performing feature dimensionality reduction processing on the data features to obtain low-dimensional data features comprises:利用主成分分析法,对所述数据特征进行特征降维处理,得到低维度数据特征。The principal component analysis method is used to perform feature dimensionality reduction processing on the data features to obtain low-dimensional data features.7.一种电力数据后门检测装置,其特征在于,所述装置包括:7. A power data backdoor detection device, characterized in that the device comprises:特征提取模块,用于将待检测电力数据输入到特征提取网络中,得到所述待检测电力数据的数据特征;所述特征提取网络中包括多个隐藏层;A feature extraction module, used for inputting the power data to be detected into a feature extraction network to obtain data features of the power data to be detected; the feature extraction network includes a plurality of hidden layers;特征降维模块,用于对所述数据特征进行特征降维处理,得到低维度数据特征;A feature dimension reduction module is used to perform feature dimension reduction processing on the data features to obtain low-dimensional data features;确定模块,用于确定所述低维度数据特征和预先确定的干净电力数据集对应的多个类别簇分别的基准聚类中心之间的第一欧式距离;A determination module, configured to determine a first Euclidean distance between the low-dimensional data feature and respective benchmark cluster centers of a plurality of category clusters corresponding to a predetermined clean power data set;所述确定模块,还用于基于多个所述第一欧式距离和预先确定的每个所述类别簇的距离约束,确定所述待检测电力数据的检测结果;所述检测结果为干净电力数据,或中毒电力数据。The determination module is further used to determine the detection result of the power data to be detected based on the plurality of the first Euclidean distances and the predetermined distance constraints of each of the category clusters; the detection result is clean power data or poisoned power data.8.一种计算机设备,其特征在于,包括存储器和处理器,所述存储器存储有计算机程序,所述处理器执行所述计算机程序时实现权利要求1至6中任一项所述的方法的步骤。8. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the methods of claims 1 to 6 when executing the computer program.9.一种计算机可读存储介质,其上存储有计算机程序,其特征在于,所述计算机程序被处理器执行时实现如权利要求1至6任一项所述的方法的步骤。9. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.10.一种计算机程序产品,包括计算机程序,其特征在于,所述计算机程序被处理器执行时实现如权利要求1至6任一项所述的方法的步骤。10. A computer program product, comprising a computer program, characterized in that when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.
CN202411759788.4A2024-12-032024-12-03 Power data backdoor detection method, device, equipment, storage medium and program productActiveCN119598453B (en)

Priority Applications (1)

Application NumberPriority DateFiling DateTitle
CN202411759788.4ACN119598453B (en)2024-12-032024-12-03 Power data backdoor detection method, device, equipment, storage medium and program product

Applications Claiming Priority (1)

Application NumberPriority DateFiling DateTitle
CN202411759788.4ACN119598453B (en)2024-12-032024-12-03 Power data backdoor detection method, device, equipment, storage medium and program product

Publications (2)

Publication NumberPublication Date
CN119598453Atrue CN119598453A (en)2025-03-11
CN119598453B CN119598453B (en)2025-09-05

Family

ID=94828840

Family Applications (1)

Application NumberTitlePriority DateFiling Date
CN202411759788.4AActiveCN119598453B (en)2024-12-032024-12-03 Power data backdoor detection method, device, equipment, storage medium and program product

Country Status (1)

CountryLink
CN (1)CN119598453B (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
US8195734B1 (en)*2006-11-272012-06-05The Research Foundation Of State University Of New YorkCombining multiple clusterings by soft correspondence
CN109978070A (en)*2019-04-032019-07-05北京市天元网络技术股份有限公司A kind of improved K-means rejecting outliers method and device
US20200152320A1 (en)*2018-11-122020-05-14Roche Molecular Systems, Inc.Medical treatment metric modelling based on machine learning
CN112765142A (en)*2021-01-182021-05-07北京易莱信科技有限公司Abnormal data diagnosis method and system based on two-step clustering algorithm
CN115270957A (en)*2022-07-252022-11-01广东电网有限责任公司Data processing method and device, electronic equipment and storage medium
US20240303330A1 (en)*2021-01-202024-09-12Nippon Telegraph And Telephone CorporationGeneration device, generation method, and generation program

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
US8195734B1 (en)*2006-11-272012-06-05The Research Foundation Of State University Of New YorkCombining multiple clusterings by soft correspondence
US20200152320A1 (en)*2018-11-122020-05-14Roche Molecular Systems, Inc.Medical treatment metric modelling based on machine learning
CN109978070A (en)*2019-04-032019-07-05北京市天元网络技术股份有限公司A kind of improved K-means rejecting outliers method and device
CN112765142A (en)*2021-01-182021-05-07北京易莱信科技有限公司Abnormal data diagnosis method and system based on two-step clustering algorithm
US20240303330A1 (en)*2021-01-202024-09-12Nippon Telegraph And Telephone CorporationGeneration device, generation method, and generation program
CN115270957A (en)*2022-07-252022-11-01广东电网有限责任公司Data processing method and device, electronic equipment and storage medium

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
TRINUGI WIRA HARJANTI等: "Classification of Mint Leaf Types Using Euclidean Distance and K-Means Clustering with Shape and Texture Feature Extraction", pages 1 - 10, Retrieved from the Internet <URL:《网页在线公开:https://jurnal.buddhidharma.ac.id/index.php/te/article/view/940》>*
陶汉等: "一种有效的高维数据分类算法", 《电脑知识与技术》, vol. 13, no. 5, 5 May 2017 (2017-05-05), pages 5 - 6*

Also Published As

Publication numberPublication date
CN119598453B (en)2025-09-05

Similar Documents

PublicationPublication DateTitle
Liu et al.CNID: research of network intrusion detection based on convolutional neural network
CN119484065B (en)Network security situation awareness method based on deep neural network
CN108564009A (en)A kind of improvement characteristic evaluation method based on mutual information
CN115409111A (en)Training method of named entity recognition model and named entity recognition method
CN115396212A (en)Training method and device for detection model, computer equipment and storage medium
CN117473993A (en)Virtual resource acquisition behavior prediction method, device, equipment and storage medium
CN118364317A (en)Sample expansion method, sample expansion device, computer equipment and readable storage medium
CN116451093A (en) Training method of circuit fault analysis model and circuit fault analysis method
CN114881158B (en) Missing value filling method, device and computer equipment based on random forest
CN119598453B (en) Power data backdoor detection method, device, equipment, storage medium and program product
CN115470526B (en) Method and device for processing anti-attack data based on black box model
CN117251777A (en)Data processing method, device, computer equipment and storage medium
CN114418060B (en) Identity-preserving adversarial training method, device, and medium based on graph representation learning
CN115982598A (en)Electricity stealing detection method, apparatus, device, medium, and program product
CN119397417B (en) Model backdoor reprogramming defense method, device, computer equipment and medium
Zhou et al.On Learning Label Noise Robust Networks via Regularization: A Topological View
CN119600368B (en) Training method, device, equipment and storage medium for power adversarial sample image recognition model
CN115146258B (en)Request processing method and device, storage medium and electronic equipment
CN119577260B (en)Multidimensional rumor detection method based on graph attention mechanism and self-supervision learning
CN119397417A (en) Model backdoor reprogramming defense method, device, computer equipment and medium
Yang et al.Feature selection based on network maximal correlation
CN119647598A (en) Large model reasoning acceleration method, device, computer equipment, readable storage medium and program product
CN119668993A (en) Log anomaly detection method, device, computer equipment, readable storage medium and program product
CN117459577A (en)Virtual article information pushing method, device, computer equipment and storage medium
CN116702138A (en)Neural network model training method for filtering false association of software and storage medium

Legal Events

DateCodeTitleDescription
PB01Publication
PB01Publication
SE01Entry into force of request for substantive examination
SE01Entry into force of request for substantive examination
GR01Patent grant
GR01Patent grant

[8]ページ先頭

©2009-2025 Movatter.jp